多任务检测网络下车道线宽度测距方法  

Ranging method based on lane width in multi-task detection network

作　　者：刘军[1] 陈辰 李汉冰 许多 LIU Jun;CHEN Chen;LI Hanbing;XU Duo(School of Automobile And Traffic Engineering, Jiangsu University, Zhenjiang 212013, China)

机构地区：[1]江苏大学汽车与交通工程学院,江苏镇江212013

出　　处：《重庆理工大学学报（自然科学）》2022年第6期62-71,共10页Journal of Chongqing University of Technology：Natural Science

基　　金：国家自然科学基金项目(51275212)。

摘　　要：针对复杂交通场景下单个模型无法同时检测车道线和车辆,以及车道线检测实时性较差的问题,提出了一种新的多任务检测网络模型,该模型可以同时检测车辆和车道线,无需任何先验信息。同时,在多任务检测网络模型下修正现有的基于车道宽度进行测距的方法:对现有方法在弯道下测距产生的误差进行分析后,提出一种弯道下基于车道线斜率的车道横向宽度的计算模型,在假设车道线是同心圆模型前提下,修正车道横向宽度计算公式后推导出前方车辆纵向距离的测距模型,车道线曲率为0.01时,对50 m范围内的车辆实施修正后的测距方法,测距误差小于3%;在车道线曲率小于0.005时,对100 m范围内的车辆实施修正后的测距方法,测距误差小于1%,相较于现有方法,修正后测距方法的测距精度显著提高。最后通过KITTI数据集验证在多任务检测网络下实施修正后的测距方法的效果,结果表明:多任务检测网络对车道线的近点拟合效果较好,且在有遮挡的情况下仍然可以预测出遮挡部分的车道线,鲁棒性较好,修正后的测距方法的平均测距误差小于5%,测距精度明显提高。To solve the problem that a single model can not detect both lanes and vehicles in complex traffic scenarios,and the real-time performance of lane line detection is poor,a new multi-task detection network model is proposed,which can detect both vehicles and lanes simultaneously without any prior information.At the same time,the existing methods based on lane width are corrected under the multi-task detection network model:After analyzing the errors caused by the existing methods in under-bend ranging,a calculation model of lane lateral width under-bend based on Lane slope is presented,assuming that the lane line is a concentric circle model.When the curvature of the lane is 0.01,the corrected ranging method is applied to vehicles within 50 meters,and the ranging error is less than 3%.When the curvature of the lane line is less than 0.005,the distance measuring error of the corrected method is less than 1%for vehicles within 100 meters.Compared with the existing methods,the corrected distance measuring method improves the distance measuring accuracy significantly.Finally,the KITTI dataset is used to verify the effectiveness of the modified ranging method under the multitask detection network proposed in this paper.The results show that the multitask detection network has a good fitting effect on the near points of the lane lines and can still predict the lane lines in the obscured part under the condition of occlusion.The robustness is good.The average ranging error of the modified ranging method is less than 5%.The ranging accuracy is improved significantly.

关 键 词：多任务检测网络 车辆检测 车道线 误差分析 测距模型修正 

分 类 号：U491.6[交通运输工程—交通运输规划与管理]